A methodology for granularity-based control of parallelism in logic programs

Several types of parallelism can be exploited in logic programs while preserving correctness and efficiency, i.e. ensuring that the parallel execution obtains the same results as the sequential one and the amount of work performed is not greater. However, such results do not take into account a number of overheads which appear in practice, such as process creation and scheduling, which can induce a slow-down, or, at least, limit speedup, if they are not controlled in some way. This paper describes a methodology whereby the granularity of parallel tasks, i.e. the work available under them, is efficiently estimated and used to limit parallelism so that the effect of such overheads is controlled. The run-time overhead associated with the approach is usually quite small, since as much work is done at compile time as possible. Also,a number of run-time optimizations are proposed. Moreover, a static analysis of the overhead associated with the granularity control process is performed in order to decide its convenience. The performance improvements resulting from the incorporation of grain size control are shown to be quite good, specially for systems with medium to large parallel execution overheads.

Automatic compile-time parallelization of logic programs for restricted, goal-level, independent and-parallelism.

A framework for the automatic parallelization of (constraint) logic programs is proposed and proved correct. Intuitively, the parallelization process replaces conjunctions of literals with parallel expressions. Such expressions trigger at run-time the exploitation of restricted, goal-level, independent and-parallelism. The parallelization process performs two steps. The first one builds a conditional dependency graph (which can be implified using compile-time analysis information), while the second transforms the resulting graph into linear conditional expressions, the parallel expressions of the &-Prolog language. Several heuristic algorithms for the latter ("annotation") process are proposed and proved correct. Algorithms are also given which determine if there is any loss of parallelism in the linearization process with respect to a proposed notion of maximal parallelism. Finally, a system is presented which implements the proposed approach. The performance of the different annotation algorithms is compared experimentally in this system by studying the time spent in parallelization and the effectiveness of the results in terms of speedups.

Relating data—parallelism and (and—) parallelism in logic programs

Much work has been done in the áreas of and-parallelism and data parallelism in Logic Programs. Such work has proceeded to a certain extent in an independent fashion. Both types of parallelism offer advantages and disadvantages. Traditional (and-) parallel models offer generality, being able to exploit parallelism in a large class of programs (including that exploited by data parallelism techniques). Data parallelism techniques on the other hand offer increased performance for a restricted class of programs. The thesis of this paper is that these two forms of parallelism are not fundamentally different and that relating them opens the possibility of obtaining the advantages of both within the same system. Some relevant issues are discussed and solutions proposed. The discussion is illustrated through visualizations of actual parallel executions implementing the ideas proposed.

High-level characteristics of or-and independent and-parallelism in prolog

Although studies of a number of parallel implementations of logic programming languages are now available, their results are difficult to interpret due to the multiplicity of factors involved, the effect of each of which is difficult to sepárate. In this paper we present the results of a high-level simulation study of or- and independent and-parallelism with a wide selection of Prolog programs that aims to determine the intrinsic amount of parallelism, independently of implementation factors, thus facilitating this separation. We expect this study will be instrumental in better understanding and comparing results from actual implementations, as shown by some examples provided in the paper. In addition, the paper examines some of the issues and tradeoffs associated with the combination of and- and or-parallelism and proposes reasonable solutions based on the simulation data obtained.

The &-prolog system: Exploiting independent and-parallelism

The &-Prolog system, a practical implementation of a parallel execution niodel for Prolog exploiting strict and non-strict independent and-parallelism, is described. Both automatic and manual parallelization of programs is supported. This description includes a summary of the system's language and architecture, some details of its execution model (based on the RAP-WAM model), and data on its performance on sequential workstations and shared memory multiprocessors, which is compared to that of current Prolog systems. The results to date show significant speed advantages over state-of-the-art sequential systems.

Strict and non-strict independent and-parallelism in logic programs: Correctness, efficiency, and compile-time conditions.

This paper presents some fundamental properties of independent and-parallelism and extends its applicability by enlarging the class of goals eligible for parallel execution. A simple model of (independent) and-parallel execution is proposed and issues of correctness and efficiency discussed in the light of this model. Two conditions, "strict" and "non-strict" independence, are defined and then proved sufficient to ensure correctness and efñciency of parallel execution: if goals which meet these conditions are executed in parallel the solutions obtained are the same as those produced by standard sequential execution. Also, in absence of failure, the parallel proof procedure does not genérate any additional work (with respect to standard SLD-resolution) while the actual execution time is reduced. Finally, in case of failure of any of the goals no slow down will occur. For strict independence the results are shown to hold independently of whether the parallel goals execute in the same environment or in sepárate environments. In addition, a formal basis is given for the automatic compile-time generation of independent and-parallelism: compile-time conditions to efficiently check goal independence at run-time are proposed and proved sufficient. Also, rules are given for constructing simpler conditions if information regarding the binding context of the goals to be executed in parallel is available to the compiler.

Annotation algorithms for unrestricted independent and-parallelism in logic programs

We present two new algorithms which perform automatic parallelization via source-to-source transformations. The objective is to exploit goal-level, unrestricted independent and-parallelism. The proposed algorithms use as targets new parallel execution primitives which are simpler and more flexible than the well-known &/2 parallel operator. This makes it possible to genérate better parallel expressions by exposing more potential parallelism among the literals of a clause than is possible with &/2. The difference between the two algorithms stems from whether the order of the solutions obtained is preserved or not. We also report on a preliminary evaluation of an implementation of our approach. We compare the performance obtained to that of previous annotation algorithms and show that relevant improvements can be obtained.

Towards independent and-parallelism in CLP

In this paper we propose a complete scheme for automatic exploitation of independent and-parallelism in CLP programs. We first discuss the new problems involved because of the different properties of the independence notions applicable to CLP. We then show how independence can be derived from a number of standard analysis domains for CLP. Finally, we perform a preliminary evaluation of the efficiency, accuracy, and effectiveness of the approach by implementing a parallehzing compiler for CLP based on the proposed ideas and applying it on a number of CLP benchmarks.

Towards granularity based control of parallelism in logic programs

Several types of parallelism can be exploited in logic programs while preserving correctness and efficiency, i.e. ensuring that the parallel execution obtains the same results as the sequential one and the amount of work performed is not greater. However, such results do not take into account a number of overheads which appear in practice, such as process creation and scheduling, which can induce a slow-down, or, at least, limit speedup, if they are not controlled in some way. This paper describes a methodology whereby the granularity of parallel tasks, i.e. the work available under them, is efficiently estimated and used to limit parallelism so that the effect of such overheads is controlled. The run-time overhead associated with the approach is usually quite small, since as much work is done at compile time as possible. Also, a number of run-time optimizations are proposed. Moreover, a static analysis of the overhead associated with the granularity control process is performed in order to decide its convenience. The performance improvements resulting from the incorporation of grain size control are shown to be quite good, specially for systems with médium to large parallel execution overheads.

Extracting non-strict independent and-parallelism using sharing and freeness information

Logic programming systems which exploit and-parallelism among non-deterministic goals rely on notions of independence among those goals in order to ensure certain efficiency properties. "Non-strict" independence (NSI) is a more relaxed notion than the traditional notion of "strict" independence (SI) which still ensures the relevant efficiency properties and can allow considerable more parallelism than SI. However, all compilation technology developed to date has been based on SI, because of the intrinsic complexity of exploiting NSI. This is related to the fact that NSI cannot be determined "a priori" as SI. This paper filis this gap by developing a technique for compile-time detection and annotation of NSI. It also proposes algorithms for combined compiletime/ run-time detection, presenting novel run-time checks for this type of parallelism. Also, a transformation procedure to eliminate shared variables among parallel goals is presented, aimed at performing as much work as possible at compile-time. The approach is based on the knowledge of certain properties regarding the run-time instantiations of program variables —sharing and freeness— for which compile-time technology is available, with new approaches being currently proposed. Thus, the paper does not deal with the analysis itself, but rather with how the analysis results can be used to parallelize programs.

Automatic exploitation of non-determinate independent and-parallelism in the basic andorra model

Andorra-I is the first implementation of a language based on the Andorra Principie, which states that determinate goals can (and shonld) be run before other goals, and even in a parallel fashion. This principie has materialized in a framework called the Basic Andorra model, which allows or-parallelism as well as (dependent) and-parallelism for determinate goals. In this report we show that it is possible to further extend this model in order to allow general independent and-parallelism for nondeterminate goals, withont greatly modifying the underlying implementation machinery. A simple an easy way to realize such an extensión is to make each (nondeterminate) independent goal determinate, by using a special "bagof" constract. We also show that this can be achieved antomatically by compile-time translation from original Prolog programs. A transformation that fulfüls this objective and which can be easily antomated is presented in this report.

A simulation study of or- and independent and-parallelism

Although studies of a number of parallel implementations of logic programming languages are now available, the results are difficult to interpret due to the multiplicity of factors involved, the effect of each of which is difficult to sepárate. In this paper we present the results of a highlevel simulation study of or- and independent and-parallelism with a wide selection of Prolog programs that aims to facilítate this separation. We hope this study will be instrumental in better understanding and comparing results from actual implementations, as shown by an example in the paper. In addition, the paper examines some of the issues and tradeoffs associated with the combination of and- and or-parallelism and proposes reasonable solutions based on the simulation data.

IDIOM: Integrating dependent and-, independent and-, and or-parallelism

Independent and-parallelism, dependent and-parallelism and or-parallelism are the three main forms of implicit parallelism present in logic programs. In this paper we present a model, IDIOM, which exploits all three forms of parallelism in a single framework. IDIOM is based on a combination of the Basic Andorra Model and the Extended And-Or Tree Model. Our model supports both Prolog as well as the fíat concurrent logic languages. We discuss the issues that arise in combining the three forms of parallelism, and our solutions to them. We also present an implementation scheme, based on binding arrays, for implementing IDIOM.

The CDG, UDG, and MEL methods for automatic compile-time parallelization of logic programs for independent and-parallelism

There has been significant interest in parallel execution models for logic programs which exploit Independent And-Parallelism (IAP). In these models, it is necessary to determine which goals are independent and therefore eligible for parallel execution and which goals have to wait for which others during execution. Although this can be done at run-time, it can imply a very heavy overhead. In this paper, we present three algorithms for automatic compiletime parallelization of logic programs using IAP. This is done by converting a clause into a graph-based computational form and then transforming this graph into linear expressions based on &-Prolog, a language for IAP. We also present an algorithm which, given a clause, determines if there is any loss of parallelism due to linearization, for the case in which only unconditional parallelism is desired. Finally, the performance of these annotation algorithms is discussed for some benchmark programs.

Non-strict independent and-parallelism

This paper presents and develops a generalized concept of Non-Strict Independent And Parallelism (NSIAP). NSIAP extends the applicability of Independent And- Parallelism (IAP) by enlarging the class of goals which are eligible for parallel execution. At the same time it maintains IAP's ability to run non-deterministic goals in parallel and to preserve the computational complexity expected in the execution of the program by the programmer. First, a parallel execution framework is defined and some fundamental correctness results, in the sense of equivalence of solutions with the sequential model, are discussed for this framework. The issue of efficiency is then considered. Two new definitions of NSI are given for the cases of puré and impure goals respectively and efficiency results are provided for programs parallelized under these definitions which include treatment of the case of goal failure: not only is reduction of execution time guaranteed (modulo run-time overheads) in the absence of failure but it is also shown that in the worst case of failure no speed-down will occur. In addition to applying to NSI, these results carry over and complete previous results shown in the context of IAP which did not deal with the case of goal failure. Finally, some practical examples of the application of the NSIAP concept to the parallelization of a set of programs are presented and performance results, showing the advantage of using NSI, are given.